The catalogue/Replicate
Replicate
app.replicate
Run any AI model with async predictions and job management
Version1.0.0
Categoryai
AuthAPI key
Tools5
Transportstdio
LicenceMIT
01
What it does
The app, in its own words.
Run any AI model on Replicate. Discover models, start async predictions, poll for results, and manage running jobs.
02
Tools
5 tools the agent can call.
Every one is declared with typed parameters and returns structured results, so the host can render them instead of narrating them.
- cancel_predictionCancel a running or queued prediction by ID.
- get_predictionGet the status and result of a prediction by ID. Poll this after run until status is "succeeded" or "failed". When succeeded, the output field contains the result (image URL, video URL, etc.).
- list_modelsDiscover models available on Replicate and their exact input parameters. Three modes: (1) collection: list models in a category (e.g. "video-generation", "image-to-video", "text-to-image", "image-to-image", "audio-generation"). (2) query: search models by name or description. (3) model: get the full input/output schema for a specific model in "owner/name" format (e.g. "google/veo-3.1") — use this before run to know exactly what parameters to pass.
- list_predictionsList recent predictions with their status. Useful to track ongoing or past generations.
- runRun any model on Replicate. ALWAYS asynchronous — returns immediately with a prediction ID and status "starting". Then use get_prediction to poll until done. Use list_models first to discover the model ID and its exact input parameters.
03
Auth and permissions
What it asks for.
Declared in the manifest, in advance. The app cannot reach anything it did not ask for here.
System secrets
API_TOKEN
Reading further
The manifest and the source for Replicate live in the monorepo, alongside the other apps.